AsyncInputStream: if input-stream.read returns empty, try again.

[pchickey]

For details: bytecodealliance/wasmtime#9667 (comment)

We can release this as 0.5.1.

[sunfishcode]

Another option, instead of looping, would be to return Err(std::io::Error::from(std::io::ErrorKind::Interrupted)). The Interrupted error code is the documented value for Rust read functions to return to indicate that zero bytes were received. Rust code doing reads should already be checking for Interrupted and using a loop (eg. like this). This approach would avoid having a redundant loop here, and would give control back to user code sooner in case their loop has other things to do while they're waiting for actual data to arrive.

[pchickey]

That would be the correct behavior for synchronous code. However, in this async context, the ready().await in this loop is effectively giving control back to the user's code to do other things while it waits, including the possibility of dropping the read Future to cancel the operation. So, in async, I think its this library's job to ensure the user never sees an ErrorKind::Interrupted.

[sunfishcode]

Investigating more, I found that it's common for async Read traits to retry on ErrorKind::Interrupted, such as futures_io::AsyncRead does. So that sounds like the thing to do here too.